Skip to content
All work

Rayna Tours · 2025–2026

A governed home for an AI agent service

An AI platform strategy for a travel business, and hosting for its new AI agent service on the existing AWS platform, with private data stores and the same guardrails as every other workload.

Diagram of the AI agent service on AWSRequests reach the AI agent service, a container service on the existing AWS platform, through the shared load balancer. The service uses a private PostgreSQL database and a private cache. It has the same guardrails as every other workload: keyless pipelines, private networking and explicit egress rules.RequestsExisting AWS platformLoad balancerSharedAI agent serviceContainer servicePostgreSQLPrivateCachePrivateSame guardrails as every workloadKeyless pipelines · Private networking · Explicit egress rules
The AI service needed no special path to production: it goes through the same reviewed pipeline and guardrails as everything else.

The situation

Rayna Tours planned to use AI in customer service and operations: chat assistance, recommendations, itinerary generation and document processing. The business first needed to choose its platforms, and the engineering team needed a safe place to run an AI agent service.

What we did

  • Wrote an AI platform decision guide that compared Azure AI, Google Cloud AI, the OpenAI API and AWS for the tourism use cases. It set clear principles: treat AI services as stateless APIs, avoid early lock-in, centralise prompt versioning and logging, and enforce data protection, rate limits and auditability.
  • Built the hosting for the AI agent service on the existing AWS platform: a container service behind the shared load balancer, a private PostgreSQL database and a private cache, wired in through the same parameter contracts as every other service.
  • Kept the same guardrails as the rest of the estate: keyless pipelines, private networking and explicit egress rules.

Why it matters

The AI service did not need a special path to production. It went through the same reviewed pipeline and guardrails as everything else. That is the point: AI workloads should inherit the platform’s controls, not bypass them.

Services used

Not sure where to start?

Book a free 30-minute call. We learn what you need and tell you honestly whether and how we can help. There is no obligation.